Add CONFIGURE parameter to specify NVIDIA cards generation upon config

[nlabrad]

So it’s a known issue that the config file currently has “compute_30” and that if you have an NVIDIA card that won’t work.

The compute number changes depending on what generation of card you’re running. For example, my GTX 1070 is compute_61 or sm_61. The 20x0 are 70 and so on. What is for sure wrong is compute_30. I think it’s so old that it’s not supported anymore.

I don’t know how to probe the system to find out what card is installed and pick the number accordingly, so I was wondering if we could add a —with-NVIDIA=Pascal/etc option to the configure script to select the correct number and avoid this issue.

I want to pick this up but I wanted some input first to know if this is the way to go

[msuchard]

My suggestion would be to remove required compute level specification. Current nvcccompilers all support by default double-precision.

So, please make a pull request and I'll merge.

Also, I personally have switched over to cmake (in the hmc-clock branch) for current beagle development and will merge that in when finished (a couple of months).

[alanorth]

For what it's worth, I'm doing this for an NVIDIA Tesla V100 that supports compute capability 7.0:

$ sed -i 's/compute_30/compute_70/' configure.ac
$ ./autogen.sh
$ ./configure
$ make